簡易檢索 / 詳目顯示

回結果列表
研究生： 陳奕萍
Chen, Yi-Ping
論文名稱： 生成式增強圖神經網路之少樣本人機假訊息偵測
Few-shot Human-Machine Fake News Detection with AI-generated Text-Augmented Graph Neural Networks
指導教授： 李政德
Li, Cheng-Te
學位類別： 碩士
Master
系所名稱： 電機資訊學院 - 資訊工程學系
Department of Computer Science and Information Engineering
論文出版年： 2025
畢業學年度： 113
語文別： 英文
論文頁數： 78
中文關鍵詞： 假新聞檢測 、大型語言模型 、少樣本學習 、構構圖神經網路 、人工智慧生成內容
外文關鍵詞： Fake news detection, Large language models, Few-shot learning, Heterogeneous graph neural networks, AI-generated content
相關次數： 點閱：72  下載：0 
分享至:
查詢本校圖書館目錄 查詢臺灣博碩士論文知識加值系統 勘誤回報

    • 隨著大型語言模型(Large Language Models, LLMs)的快速發展，假新聞的產生和傳播形式發生了根本性的改變。LLM不僅能快速生成具有說服力的文本內容，還可能產生比人工撰寫更具可信度的假新聞。這種轉變使得傳統的假新聞檢測方法面臨新的挑戰。為了彌補現有數據集與真實新聞生態之間的差距，我們構建了一個綜合數據集AuthGenNews，通過整合FakeNewsNet和CoAID基準數據集，並引入多元且專業導向的 LLM 生成策略。該數據集包含人類撰寫及AI 生成的新聞，更貼近現代新聞來源的多樣性。
    本篇論文提出了一個創新框架，結合LLM的生成能力和圖神經網路的結構學習優勢，以解決假新聞檢測中的早期檢測(當新聞文章剛發布時即進行檢測)、資料稀缺(獲取足夠標註數據進行訓練的困難性)和人機混合內容識別等關鍵挑戰。具體而言，我們創新地運用AI生成文本增強特徵表示，並設計異構圖神經網路進行小樣本學習，以最少的訓練資料提高檢測準確性。
    實驗結果表明，在極端少樣本情況下(5%訓練數據)，該框架在AuthGenNews-FakeNewsNet數據集上達到了75.0%的準確率，在AuthGenNews-CoAID數據集上則達到了93.7%的準確率。本研究透過結合LLM 與圖神經網路的優勢，成功克服了現有方法在極端少樣本環境下的局限性，顯著提升了僅依賴新聞文章的假新聞檢測能力。為現代新聞生態中的早期假新聞檢測提供了一種創新的解決方案。

    The rapid advancement of Large Language Models (LLMs) has fundamentally transformed the generation and dissemination of fake news. LLMs can not only quickly generate persuasive textual content but also potentially produce fake news that is more credible than human-written content. This transformation poses new challenges to traditional fake news detection methods. To bridge the gap between existing datasets and real-world news ecosystems, we construct AuthGenNews, a comprehensive dataset that integrates FakeNewsNet and CoAID benchmark datasets while incorporating diverse, profession-oriented LLM generation strategies. This dataset includes both human-written and AI-generated news, providing a more realistic representation of modern news sources.
    This study propose an innovative framework that combines the generative capabilities of LLMs and the structural learning advantages of Graph Neural Networks (GNNs) to address key challenges such as early detection (detecting fake news as soon as a news article is published) and data scarcity (the difficulty of obtaining sufficient labeled data for training), and human-AI hybrid content identification in fake news detection. The framework enriches feature representation using AI-generated text and employs a heterogeneous GNN for few-shot learning, improving detection accuracy with minimal training data.
    Experimental results demonstrate that under extreme few-shot conditions (5% training data), the framework achieves 75.0% accuracy on the AuthGenNews-FakeNewsNet dataset and 93.7% accuracy on the AuthGenNews-CoAID dataset. By integrating the strengths of Large Language Models (LLM) and Graph Neural Networks (GNN), this study successfully overcomes the limitations of existing methods in extreme few-shot settings, significantly enhancing fake news detection using only news articles. This research provides an innovative solution for early-stage fake news detection in modern news ecosystems.

    摘要 I Abstract II Table of Contents III List of Tables VI List of Figures VII Chapter1 Introduction 1 Chapter2 Related Work 5 2.1 Fake News Detection 5 2.1.1 LLM-Based Methods 5 2.1.2 GNN-Based Methods 6 2.2 LLM-Generated Fake News and Datasets 7 2.3 Summary 9 Chapter3 Problem Formulation 11 3.1 Basic Definitions 11 3.2 Few-shot Learning Setting 11 3.3 Problem Objective 12 Chapter4 Methods 13 4.1 Framework Overview 14 4.2 LLM-based Feature Generation 14 4.3 Heterogeneous Graph Construction 18 4.3.1 Feature Extraction with Language Models 18 4.3.1.1 Semantic Feature Extraction 18 4.3.1.2 Sentiment Feature Extraction 20 4.3.2 Heterogeneous Graph Construction 22 4.3.2.1 Meta-path in Heterogeneous Graph 23 4.4 Few-shot Multi-task HGNN Learning 26 4.4.1 Model Architecture 27 4.4.1.1 Heterogeneous Graph Convolution 27 4.4.1.2 Heterogeneous Graph Neural Network Architecture 28 4.4.1.3 Extended Architecture for Additional Text Augmentation 33 4.4.2 Multi-task Loss Function 34 Chapter5 Experimental Results 36 5.1 Dataset 36 5.1.1 FakeNewsNet Dataset 36 5.1.2 CoAID Dataset 37 5.1.3 Dataset Generation 37 5.1.4 LLM-Generated Methods 39 5.2 Experimental Setup 42 5.2.1 Baseline Methods 42 5.2.2 Evaluation Metrics 43 5.2.3 Inductive Learning Setup 43 5.3 Performance Analysis 44 5.3.1 Fake News Detection Performance 44 5.3.2 Fake News Detections Performance on Original Datasets 48 5.3.3 Human-AI Text Classification Performance 49 5.4 Hyperparameter Study 52 5.4.1 Comment Length Analysis 52 5.4.2 Loss Weight Analysis 54 5.4.3 Augmentation Type Selection 56 5.5 Ablation study 57 5.5.1 Framework Component 57 5.5.2 Training Strategies 58 5.5.3 Text-Augmentation Strategies 61 Chapter6 Conclusion 64 6.1 Research Contributions 64 6.2 Experimental Results 65 6.3 Future Research Directions 66 References 67

    [1] MARTINO, Giovanni Da San, et al. A survey on computational propaganda detection. arXiv preprint arXiv:2007.08024, 2020.
    [2] Ray, Partha Pratim. "ChatGPT: A comprehensive review on background, applications, key challenges, bias, ethics, limitations and future scope. Internet of Things and Cyber-Physical Systems, 3, 121–154." URL https://doi. org/10.1016/j. iotcps 3 (2023).
    [3] Shu, Kai, et al. "Fakenewsnet: A data repository with news content, social context, and spatiotemporal information for studying fake news on social media." Big data 8.3 (2020): 171-188.
    [4] Cui, Limeng, and Dongwon Lee. "Coaid: Covid-19 healthcare misinformation dataset." arXiv preprint arXiv:2006.00885 (2020).
    [5] Achiam, Josh, et al. "Gpt-4 technical report." arXiv preprint arXiv:2303.08774 (2023).
    [6] Zellers, Rowan, et al. "Defending against neural fake news." Advances in neural information processing systems 32 (2019).
    [7] Pennycook, Gordon, and David G. Rand. "Fighting misinformation on social media using crowdsourced judgments of news source quality." Proceedings of the National Academy of Sciences 116.7 (2019): 2521-2526.
    [8] Chen, Canyu, and Kai Shu. "Can llm-generated misinformation be detected?." arXiv preprint arXiv:2309.13788 (2023).
    [9] Vykopal, Ivan, et al. "Disinformation capabilities of large language models." arXiv preprint arXiv:2311.08838 (2023).
    [10] Jiang, Bohan, et al. "Disinformation detection: An evolving challenge in the age of llms." Proceedings of the 2024 SIAM International Conference on Data Mining (SDM). Society for Industrial and Applied Mathematics, 2024.
    [11] Lucas, Jason S., et al. "Fighting Fire with Fire: The Dual Role of LLMs in Crafting and Detecting Elusive Disinformation." The 2023 Conference on Empirical Methods in Natural Language Processing.
    [12] Wang, Lionel Z., et al. "Megafake: a theory-driven dataset of fake news generated by large language models." arXiv preprint arXiv:2408.11871 (2024).
    [13] Su, Jinyan, et al. "Fake news detectors are biased against texts generated by large language models." arXiv preprint arXiv:2309.08674 (2023).
    [14] Su, Jinyan, Claire Cardie, and Preslav Nakov. "Adapting Fake News Detection to the Era of Large Language Models." Findings of the Association for Computational Linguistics: NAACL 2024. 2024.
    [15] Liu, Yinhan. "Roberta: A robustly optimized bert pretraining approach." arXiv preprint arXiv:1907.11692 364 (2019).
    [16] Kenton, Jacob Devlin Ming-Wei Chang, and Lee Kristina Toutanova. "Bert: Pre-training of deep bidirectional transformers for language understanding." Proceedings of naacL-HLT. Vol. 1. No. 2. 2019.
    [17] Sun, Yizhou, et al. "Pathsim: Meta path-based top-k similarity search in heterogeneous information networks." Proceedings of the VLDB Endowment 4.11 (2011): 992-1003.
    [18] NLP Town. "bert-base-multilingual-uncased-sentiment (Revision edd66ab) ." (2023).
    [19] Liu, Ye, et al. "Detect, investigate, judge and determine: A novel llm-based framework for few-shot fake news detection." arXiv preprint arXiv:2407.08952 (2024).
    [20] Li, Xinyi, Yongfeng Zhang, and Edward C. Malthouse. "Large Language Model Agent for Fake News Detection." arXiv preprint arXiv:2405.01593 (2024).
    [21] Huang, Yue, and Lichao Sun. "FakeGPT: fake news generation, explanation and detection of large language models." arXiv preprint arxiv:2310.05046 (2024).
    [22] Bian, Tian, et al. "Rumor detection on social media with bi-directional graph convolutional networks." Proceedings of the AAAI conference on artificial intelligence. Vol. 34. No. 01. 2020.
    [23] Zhou, Xinyi, Jindi Wu, and Reza Zafarani. ": Similarity-aware multi-modal fake news detection." Pacific-Asia Conference on knowledge discovery and data mining. Cham: Springer International Publishing, 2020.
    [24] Ren, Yuxiang, and Jiawei Zhang. "Fake news detection on news-oriented heterogeneous information networks through hierarchical graph attention." 2021 International Joint Conference on Neural Networks (IJCNN). IEEE, 2021.
    [25] Zarocostas, John. "How to fight an infodemic." The lancet 395.10225 (2020): 676.
    [26] Kreps, Sarah, R. Miles McCain, and Miles Brundage. "All the news that’s fit to fabricate: AI-generated text as a tool of media misinformation." Journal of experimental political science 9.1 (2022): 104-117.
    [27] Shu, Kai, Suhang Wang, and Huan Liu. "Beyond news contents: The role of social context for fake news detection." Proceedings of the twelfth ACM international conference on web search and data mining. 2019.
    [28] Zhou, Xinyi, and Reza Zafarani. "A survey of fake news: Fundamental theories, detection methods, and opportunities." ACM Computing Surveys (CSUR) 53.5 (2020): 1-40.
    [29] Chen, Zhikai, et al. "Label-free Node Classification on Graphs with Large Language Models (LLMs)." The Twelfth International Conference on Learning Representations. (2023).
    [30] Velickovic, Petar, et al. "Graph attention networks." stat 1050.20 (2017): 10-48550.
    [31] Kipf, Thomas N., and Max Welling. "Semi-supervised classification with graph convolutional networks." arXiv preprint arXiv:1609.02907 (2016).

    無法下載圖示 校內：2030-02-08公開
    校外：2030-02-08公開
    電子論文尚未授權公開，紙本請查館藏目錄
    QR CODE